Essays in Finance

In the first chapter of this PhD Thesis, we investigate why large cross-sections of long- short anomaly portfolios predict the market excess return. We develop an econometric model for the prices of the long and short legs of the anomalies. Using dimension reduction techniques, we show that their deviations from equilibrium predict the aggregate market return. This result holds at multiple horizons and is mostly driven by the long components of the anomaly portfolios. We interpret these findings through an asymmetric limits of arbitrage model with slow-moving capital. In the second chapter, we compare the information contained in the headlines and the full text of more than 400,000 business news articles. We show that sentiment measures extracted from the two sources are highly correlated. Using state-of-the-art machine learning methods, headline-based forecasts of macroeconomic indicators have equal or greater accuracy out-of-sample than forecasts based on the whole text. We interpret our findings through a model of news with attention costs and beauty contest elements. In the third chapter, we investigate whether measures of sentiment extracted from quarterly earnings conference-calls affect the dynamics of stock prices. Using a cross- section of publicly traded companies, we show that sentiment positively correlates with price deviations from their long-run trend, estimated via an error correction model. We document that even though sentiment does not predict future stock returns, it impacts the speed at which prices revert to equilibrium. We find asymmetric effects on overpriced and underpriced stocks.